Power equipment logic data mapping method and system, terminal and storage medium

Abstract

The invention provides a power equipment logic data mapping method and system, a terminal and a storage medium, and the method comprises the steps: constructing a semantic attribute matrix based on pre-collected power equipment physical measurement point data and label information thereof; constructing a semantic vector set and candidate logic fields based on a preset semantic analysis algorithm and the semantic attribute matrix, constructing an initial decision model through a preset decision algorithm, and performing iterative optimization on the initial decision model based on a strategy optimization algorithm to obtain a reinforcement learning strategy model, and extracting a logic data mapping action meeting a preset requirement from the candidate logic field, and executing the logic data mapping action on the to-be-processed physical measurement point of each power device to obtain logic data corresponding to the to-be-processed physical measurement point. According to the invention, the problem of poor adaptability in the mapping process from the physical measurement point to the logic field of the power equipment in the prior art is solved. According to the invention, the automatic mapping and optimization of the logic field of the physical measurement point are realized, and the adaptability of a power grid system in a complex scene is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power informatization, and in particular to a power equipment logical data mapping method, system, terminal and storage medium. BACKGROUND

[0002] With a large number of power equipment accessing the power utilization automation system and the dispatching control center, in order to realize effective interconnection and unified control among multi-source heterogeneous devices, the system needs to logically process and semantically align the physical measurement point data of various devices, and map them into structured and standardized logical fields, so as to support the rapid calling and analysis of upper-layer dispatching, monitoring and operation and maintenance functions. However, due to the current variety of power equipment, the non-uniform naming rules of manufacturers, the lack of unified coding specifications for measurement point labels, and the existence of a large number of label redundancies, semantic ambiguities and unstructured descriptions, the efficiency of the logical data mapping process is severely limited. Therefore, how to improve the adaptability of logical data mapping has become a problem to be studied.

[0003] At present, traditional data mapping methods mostly rely on manual rule configuration, regular expression matching or static template generation. Although they have certain practicability in small-scale closed systems, they show poor adaptability, weak generalization ability, insufficient scalability and easy-to-configure errors when facing large-scale heterogeneous device concurrent access, complex and ambiguous label semantics and other scenarios. In addition, traditional methods are difficult to capture deep semantic associations of labels, and cannot effectively combine device structure, operation mode and unit constraints and other multi-dimensional information for auxiliary modeling. When dealing with device labels with inconsistent semantics, unclear structure and non-standard description, the effect is particularly limited. Therefore, the existing technology has the problem of poor adaptability in the mapping process from power equipment physical measurement points to logical fields. SUMMARY

[0004] In order to solve the above problems, the present application provides a power equipment logical data mapping method, system, terminal and storage medium, which realizes automatic mapping and optimization of logical fields of physical measurement points and improves the adaptability of power grid systems in complex scenarios.

[0005] To achieve the above object, the embodiment of the present application provides a power equipment logical data mapping method, comprising: constructing a semantic attribute matrix based on pre-collected power equipment physical measurement point data and label information thereof; constructing a semantic vector set and a candidate logical field based on a preset semantic analysis algorithm and the semantic attribute matrix; constructing an initial decision model by a preset decision algorithm based on the semantic vector set and the candidate logical field, and iteratively optimizing the initial decision model based on a strategy optimization algorithm to obtain a reinforcement learning strategy model; extracting a logical data mapping action meeting a preset requirement from the candidate logical field based on the reinforcement learning strategy model, and performing the logical data mapping action on the to-be-processed physical measurement points of each power equipment to obtain the logical data corresponding to the to-be-processed physical measurement points.

[0006] The embodiment of the present application provides a power equipment logical data mapping method, physical measurement point data and label information of power equipment are collected for structured modeling, a semantic attribute matrix, a semantic vector set and a candidate logical field are obtained, a preset decision algorithm and a strategy optimization algorithm are introduced, a reinforcement learning strategy model is constructed, and it is ensured that the reinforcement learning strategy model can continuously optimize and update a mapping strategy, so that the reinforcement learning strategy model can automatically extract a logical data mapping action meeting a preset requirement from the candidate logical field, and the logical data mapping action is executed on each power equipment physical measurement point to be processed, and finally, power equipment logical data mapping is completed. Therefore, the deep meaning of the measurement point label is understood through semantic analysis, and the reinforcement learning model is constructed by using structured information, so that the logical field of the physical measurement point is automatically mapped and optimized, and the adaptability of the power grid system in a complex scene is improved.

[0007] Further, based on the pre-collected physical measurement point data and label information of the power equipment, a semantic attribute matrix is constructed, including: collecting the physical measurement point data and label information of the power equipment, and constructing an observation variable set; based on the power equipment physical attribution relationship and logical function positioning of each power equipment physical measurement point, a mapping set is constructed, and a mapping relationship corresponding to each power equipment physical measurement point is constructed; based on the mapping relationship corresponding to each power equipment physical measurement point, a numerical expression of each power equipment physical measurement point in several dimensions is constructed, and a semantic attribute matrix is obtained.

[0008] In the above scheme, the physical measurement point is formalized and modeled as an observation variable set, and the power equipment physical attribution relationship and logical function positioning of each power equipment physical measurement point are associated to obtain a mapping set and a corresponding mapping relationship, so as to construct a semantic attribute matrix that can represent the multi-dimensional semantic information of structured data. Therefore, by constructing a semantic attribute matrix, the matrix not only includes the information of the physical measurement point label itself, but also deeply integrates the structured context of the power equipment physical measurement point, so that each measurement point can be characterized in a unified and understandable high-dimensional space, providing a reliable data basis for the construction of the subsequent reinforcement learning strategy model, and helping to realize automatic mapping and optimization of the logical field of the physical measurement point, and improving the adaptability of the power grid system in a complex scene.

[0009] Further, based on the preset semantic analysis algorithm and the semantic attribute matrix, a semantic vector set and a candidate logical field are constructed, including: converting the semantic attribute matrix into a low-dimensional continuous semantic vector through the preset semantic analysis algorithm to obtain the semantic vector set; calculating the semantic distance between different physical measurement points of the power equipment or the semantic distance between the different physical measurement points of the power equipment and the preset standard logical field based on the preset similarity measurement algorithm to construct a semantic space; and preliminarily matching the physical measurement points of the power equipment and the preset standard logical field in the semantic space based on the preset search algorithm to obtain the candidate logical field.

[0010] In the above scheme, the semantic attribute matrix is converted into a low-dimensional semantic vector by using the preset semantic analysis algorithm, which provides a data basis for the construction of the subsequent reinforcement learning strategy model. Then, the semantic space is constructed by using the preset similarity measurement algorithm, and the preliminary matching of the physical measurement points of the power equipment and the preset standard logical field is performed by using the preset search algorithm, which can quickly filter out the candidate logical field most related to the physical measurement points of the power equipment from a plurality of standard logical fields. Thus, the complex global search problem is converted into a problem of searching in a highly relevant local candidate set, which provides a reliable data basis for the subsequent reinforcement learning strategy model training and guarantees the possibility of efficient self-adaptive mapping of the subsequent reinforcement learning strategy model, which helps to realize the automatic mapping and optimization of the logical field of the physical measurement point and improve the adaptability of the power grid system in a complex scenario.

[0011] Further, based on the semantic vector set and the candidate logical field, an initial decision model is constructed by using a preset decision algorithm, and the initial decision model is iteratively optimized based on a strategy optimization algorithm to obtain a reinforcement learning strategy model, including: constructing a state space based on the semantic vector set, the candidate logical field and a historical matching trajectory vector, and constructing an action space based on the candidate logical field, wherein the historical matching trajectory vector is generated when the physical measurement points of the power equipment are preliminarily matched with the preset standard logical field; calculating the semantic similarity of the physical measurement points of the power equipment and the preset standard logical field; evaluating the unit consistency, device type consistency and historical matching consistency of the physical measurement points of the power equipment and the preset standard logical field to construct a reward function; constructing the initial decision model based on the state space, the action space and the reward function; executing simulated mapping through the initial decision model based on the preset strategy optimization algorithm to generate a reward signal, and iteratively optimizing the initial decision model based on the reward signal to obtain the reinforcement learning strategy model, including: executing simulated mapping through the initial decision model based on the preset strategy optimization algorithm: obtaining the probability corresponding to the selection of the action of the action space from the state of the state space in each simulated mapping execution, generating a reward signal, and ending the simulated mapping; and based on the reward signal, iteratively optimizing the initial decision model until the reward signal meets the preset decision requirement to obtain the reinforcement learning strategy model.

[0012] In the above scheme, a state space combining the semantic vector set, the candidate logical field and the historical matching track is constructed, and an action space based on the candidate logical field is designed, and a reward function is designed to constrain the semantic similarity between the physical measurement point of the power equipment and the preset standard logical field, the unit consistency between the physical measurement point of the power equipment and the preset standard logical field, the device type consistency and the historical matching consistency, so as to construct an initial decision model, and then the initial decision model is simulated and mapped by combining a preset strategy optimization algorithm, and iterative optimization is performed according to the generated reward signal, and finally a reinforcement learning strategy model is obtained. Therefore, the static classification problem is converted into a dynamic decision problem considering multiple factors, so as to realize automatic mapping and optimization of the logical field of the physical measurement point, and improve the adaptability of the power grid system in a complex scene.

[0013] Further, based on the reinforcement learning strategy model, a logical data mapping action satisfying the preset requirement is extracted from the candidate logical field, and the logical data mapping action is performed on the to-be-processed physical measurement point of each power equipment to obtain the corresponding logical data of the to-be-processed physical measurement point, including: based on the reinforcement learning strategy model, generating logical data mapping action probability distribution data set; based on the logical data mapping action probability distribution data set, extracting the logical data mapping action satisfying the preset requirement from the candidate logical field; based on the logical data mapping action, mapping the to-be-processed physical measurement point of each power equipment to the corresponding logical field to complete the logical data mapping of the power equipment.

[0014] In the above scheme, the reinforcement learning strategy model is obtained by iterative optimization, the logical data mapping action probability distribution is obtained, and the logical data mapping action satisfying the preset requirement is screened out based on the probability distribution, so that the logical data mapping action is issued to the physical measurement point of the power equipment, and the corresponding logical data mapping action is performed on the physical measurement point of the power equipment, the to-be-processed physical measurement point of each power equipment is mapped to the corresponding logical field, and the logical data mapping of the power equipment is completed. Therefore, automatic mapping and optimization of the logical field of the physical measurement point are realized, and the adaptability of the power grid system in a complex scene is improved.

[0015] Further, each time the logical data mapping action satisfying the preset requirement is extracted from the candidate logical field based on the reinforcement learning strategy model, the logical data mapping action is performed on the to-be-processed physical measurement point of each power equipment to obtain the corresponding logical data of the to-be-processed physical measurement point, and after the step, it further includes: receiving a plurality of mapping results obtained by performing the logical data mapping action on the to-be-processed physical measurement point of each power equipment each time, and injecting the plurality of mapping results into a preset experience buffer pool at a preset step length; based on the preset experience buffer pool, incrementally learning the reinforcement learning strategy model, and adaptively optimizing the model according to the logical data mapping action probability distribution data set or the logical structure.

[0016] In the above scheme, an incremental learning mechanism is introduced, and the mapping result generated after each execution of the logical data mapping action is fed back to the preset experience buffer pool at a preset step length to adjust the logical data mapping action probability distribution data set of the reinforcement learning strategy model or fine-tune the logical structure, thereby realizing continuous optimization of the reinforcement learning strategy model, helping to realize automatic mapping and optimization of the logical field of the physical measuring point, and improving the adaptability of the power grid system in complex scenarios.

[0017] The embodiment of the present application also provides a power equipment logical data mapping system, comprising a semantic attribute construction module, a semantic analysis module, a reinforcement learning strategy model construction module and a logical data mapping module; the semantic attribute construction module is used for constructing a semantic attribute matrix based on pre-acquired power equipment physical measuring point data and label information thereof; the semantic analysis module is used for constructing a semantic vector set and a candidate logical field based on a preset semantic analysis algorithm and the semantic attribute matrix; the reinforcement learning strategy model construction module is used for constructing an initial decision model based on the semantic vector set and the candidate logical field through a preset decision algorithm, and iteratively optimizing the initial decision model based on a strategy optimization algorithm to obtain a reinforcement learning strategy model; and the logical data mapping module is used for extracting a logical data mapping action meeting a preset requirement from the candidate logical field based on the reinforcement learning strategy model, performing the logical data mapping action on each power equipment physical measuring point to be processed, and obtaining logical data corresponding to the physical measuring point to be processed.

[0018] The embodiment of the present application provides a power equipment logical data mapping system, acquires power equipment physical measuring point data and label information thereof for structured modeling, obtains a semantic attribute matrix, a semantic vector set and a candidate logical field, and then introduces a preset decision algorithm and a strategy optimization algorithm to construct a reinforcement learning strategy model, so that the reinforcement learning strategy model can continuously optimize and update the mapping strategy, so that the reinforcement learning strategy model can automatically extract a logical data mapping action meeting a preset requirement from the candidate logical field, perform the logical data mapping action on each power equipment physical measuring point to be processed, and finally complete power equipment logical data mapping. In this way, the deep meaning of the measuring point label is understood through semantic analysis, and the reinforcement learning model is constructed by using structured information, so as to realize automatic mapping and optimization of the logical field of the physical measuring point, and improve the adaptability of the power grid system in complex scenarios.

[0019] A power equipment logical data mapping terminal, the terminal comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, the processor implementing a power equipment logical data mapping method when executing the computer program.

[0020] A power equipment logical data mapping storage medium, the storage medium comprising a stored computer program, wherein the storage medium controls the device where the storage medium is located to execute a power equipment logical data mapping method when the computer program is running. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the steps of a power equipment logic data mapping method according to a certain embodiment of the present invention. Figure 2 This is a schematic diagram of the module structure of a power equipment logic data mapping system provided in a certain embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1 See Figure 1 , Figure 1 This is a flowchart illustrating the steps of a power equipment logic data mapping method according to a certain embodiment of the present invention. Figure 1 As shown, this embodiment of the invention proposes a method for mapping logical data of power equipment, including steps 101 to 104, each step of which is as follows: Step 101: Construct a semantic attribute matrix based on the pre-collected physical measurement point data of power equipment and its tag information; Step 102: Based on the preset semantic analysis algorithm and semantic attribute matrix, construct a semantic vector set and candidate logical fields; Step 103: Based on the semantic vector set and candidate logical fields, an initial decision model is constructed using a preset decision algorithm, and the initial decision model is iteratively optimized using a policy optimization algorithm to obtain a reinforcement learning policy model; Step 104: Based on the reinforcement learning policy model, extract logical data mapping actions that meet the preset requirements from the candidate logical fields, and perform logical data mapping actions on the physical measurement points to be processed for each power device to obtain the logical data corresponding to the physical measurement points to be processed.

[0024] In a specific implementation, first, the physical measurement point data of the power equipment and its label information are collected, and a semantic attribute matrix is constructed according to a plurality of structured representations, wherein the plurality of structured representations include device types, structural components, operating units and the like; then, according to a preset semantic analysis algorithm, the label, unit and evaluation range of the physical measurement point and other information are converted into a semantic vector from the semantic attribute matrix, and a candidate logical field to which the physical measurement point may correspond is further identified; then, using the semantic vector set and the candidate logical field, an initial decision model is constructed through a preset decision algorithm, in this embodiment, the mapping process of the physical measurement point to the logical field is abstracted as a Markov decision process, which is represented as the preset decision algorithm; then, a state space, an action space and a reward function are defined to construct a reinforcement learning strategy model, it is worth mentioning that in the model training process, the field mapping action of each physical measurement point is executed, and then the initial decision model is iteratively optimized according to a preset strategy optimization algorithm, and finally the reinforcement learning strategy model is obtained; finally, the trained reinforcement learning model is deployed to the actual environment of the power system, and when a new power equipment is online, the logical data mapping action is automatically performed on the to-be-processed physical measurement point of the new power equipment, so as to complete the logical data mapping of the new power equipment.

[0025] The embodiment of the present application proposes a power equipment logical data mapping method, which collects physical measurement point data of power equipment and its label information for structured modeling, obtains a semantic attribute matrix, a semantic vector set and a candidate logical field, and then introduces a preset decision algorithm and a strategy optimization algorithm to construct a reinforcement learning strategy model, which can continuously optimize and update the mapping strategy, so that the reinforcement learning strategy model can automatically extract a logical data mapping action that meets the preset requirements from the candidate logical field, and perform a logical data mapping action on each to-be-processed physical measurement point of the power equipment, and finally complete the logical data mapping of the power equipment. Thus, the deep meaning of the measurement point label is understood through semantic analysis, and the reinforcement learning model is constructed using structured information to realize automatic mapping and optimization of the logical field of the physical measurement point, and improve the adaptability of the power grid system in complex scenarios.

[0026] In a preferred scheme, based on the pre-collected physical measurement point data of the power equipment and its label information, a semantic attribute matrix is constructed, including: collecting the physical measurement point data of the power equipment and its label information, constructing an observation variable set; based on the power equipment physical attribution relationship and the logical function positioning of each power equipment physical measurement point, constructing a mapping set, and constructing the mapping relationship corresponding to each power equipment physical measurement point; based on the mapping relationship corresponding to each power equipment physical measurement point, constructing the numerical expression of each power equipment physical measurement point in several dimensions, to obtain the semantic attribute matrix.

[0027] One preferred implementation involves collecting physical measurement point data and tag information of power equipment, and formally modeling the collected physical measurement point data and tag information of power equipment as a set of observed variables M, represented as... , where each element This represents an independent physical measurement point, encompassing multiple attributes such as its associated identifier, measurement point label, measurement unit, value range, accuracy level, and time characteristics. Then, a mapping set is constructed based on the physical affiliation and logical function positioning of each power equipment physical measurement point. In this embodiment, a device type set is defined. Structural component collection and the set of operation units These correspond to various structured representations of equipment's classification attributes, functional module divisions, and control logic entities; for any physical measurement point Build mapping relationship As a set of mappings, the triples This is characterized by the physical attribution and logical functional positioning of the physical measurement point within the actual power equipment physical measurement points. It is worth noting that this mapping set not only reflects the semantic location of the physical measurement point within the power system but also provides data support for subsequent semantic analysis and logical mapping. Based on the mapping set, the numerical representation of each power equipment physical measurement point across several dimensions is abstracted into a semantic attribute matrix X, represented as... , of which Row corresponding measurement points , No. Column corresponding to the first Each semantic dimension feature Indicates the measuring point Numerical representations within this dimension include: label coding, unit category coding, device type coding, and function attribution identifiers.

[0028] In the above scheme, physical measurement points are formally modeled as a set of observed variables. A mapping set and corresponding mapping relationships are obtained by associating the physical affiliation and logical function positioning of each power equipment physical measurement point. This constructs a semantic attribute matrix that can represent the multidimensional semantic information of structured data. Thus, by constructing a semantic attribute matrix, which not only includes the information of the physical measurement point labels themselves but also deeply integrates the structured context of the power equipment physical measurement points, each measurement point can be represented in a unified and understandable high-dimensional space. This provides a reliable data foundation for the subsequent construction of reinforcement learning policy models, helps to achieve automatic mapping and optimization of the logical fields of physical measurement points, and improves the adaptability of the power grid system in complex scenarios.

[0029] In an embodiment, the semantic attribute matrix is converted into a low-dimensional continuous semantic vector set by using a preset semantic analysis algorithm, and the semantic distance between different physical measurement points of the power equipment or between the physical measurement points and the preset standard logical field is calculated based on a preset similarity measurement algorithm to construct a semantic space.

[0030] In an embodiment, the semantic attribute matrix constructed in step 101 is converted into a semantic vector set by using a preset semantic analysis algorithm. In this embodiment, the preset semantic analysis algorithm can be implemented by using a BERT embedded encoder. The specific implementation process is as follows: the label, unit, and device description of the physical measurement point are extracted from the semantic attribute matrix and input into a pre-trained model, and the output is a low-dimensional continuous semantic vector , forming a measurement point embedding set In this embodiment, V represents the semantic vector set. Then, the semantic distance between different measurement points or between the measurement points and the standard logical field is quantified by using a preset similarity measurement algorithm. In this embodiment, the preset similarity measurement algorithm can be implemented by using a semantic similarity measurement function Cosine similarity is used as the main measurement method to construct a semantic space, and the value range is [−1, 1]. The closer the value is to 1, the more similar the semantics are. The specific implementation process is as follows: first, the text and vector of the measurement point and the standard logical field are uniformly processed, including vector normalization, unit conversion to a unified standard, alias and synonym merging, and stop word cleaning. At the same time, the structural context such as the device level and the loop position, and the numerical attributes such as the rated value, the range, and the accuracy are extracted to form a comparable standardized feature set. Then, the cosine similarity of the semantic vector, the text coverage similarity based on weighted terms, the structural context similarity based on adjacency / hierarchical co-occurrence, the numerical attribute compatibility based on interval overlap and difference penalty, and the historical prior similarity based on the statistical calculation of the confirmed matching are calculated, and the comprehensive similarity is obtained by linear fusion according to the preset weight. The preset weight can be set according to the actual situation, and will not be described here. Finally, a hard consistency check is performed. The candidates with inconsistent units or device types are directly excluded. The candidates with slight inconsistencies in the range, phase, and polarity are down-weighted. Then, the double-threshold is used for hierarchical screening: directly judged as a strong match; as a candidate; rejected, and the candidates that pass the check are sorted in descending order of the comprehensive similarity, and the top-N is output as the final logical field. In this embodiment, the threshold and a threshold value Can be set according to actual conditions; in the absence of vectors or insufficient evidence, fallback to rule-based and edit distance-based guaranteed comparison to ensure availability; Finally, the preset search algorithm is used to preliminarily match the power equipment physical measuring point and the preset standard logical field in the semantic space to obtain the candidate logical field. In the embodiment, the preset search algorithm can use the K-nearest neighbor search method, and the specific execution process is as follows: first, for each to-be-mapped measuring point, the semantic distance between it and all standard logical field vectors is calculated, and the distance is sorted from small to large. The K-nearest neighbor algorithm selects the K nearest logical field samples to form the neighborhood set of the measuring point in the semantic space, thereby reflecting the most similar candidate logical relationship of the measuring point; after obtaining the K neighboring samples, the neighborhood samples can be aggregated and judged according to the similarity weighting or the simple majority principle. If multiple neighbor fields belong to the same logical category, it can be considered that this category is the most similar to the semantic of the measuring point; if the neighbors are scattered, the matching confidence of each logical field is calculated according to the similarity weight; finally, the K-nearest neighbor result is verified and modified in consistency according to the field metadata, unit, device type and measurement attribute, and the neighbors that do not meet the physical or logical constraints are weighted or removed; then the remaining candidates are reordered according to the weighted similarity, and the first several items are selected as the final candidate logical field. The K-nearest neighbor search method can significantly reduce the mapping candidate space of the semantic rough matching strategy.

[0031] In the above scheme, the semantic attribute matrix is converted into a low-dimensional semantic vector by using the preset semantic analysis algorithm, which provides a data basis for the construction of the subsequent reinforcement learning strategy model. Then, the semantic space is constructed by using the preset similarity measurement algorithm, and the preliminary matching of the power equipment physical measuring point and the preset standard logical field is performed by using the preset search algorithm, which can quickly filter out the most relevant candidate logical field from several standard logical fields. Thus, the complex global search problem is converted into a problem of searching in a highly relevant local candidate set, which provides a reliable data basis for the subsequent reinforcement learning strategy model training and guarantees the possibility of efficient adaptive mapping of the subsequent reinforcement learning strategy model, which helps to realize the automatic mapping and optimization of the logical field of the physical measuring point and improve the adaptability of the power grid system in complex scenarios.

[0032] In an embodiment, the initial decision model is constructed based on the semantic vector set and the candidate logical field by using a preset decision algorithm, and the initial decision model is iteratively optimized based on a preset strategy optimization algorithm to obtain a reinforcement learning strategy model, including: constructing a state space based on the semantic vector set, the candidate logical field and a historical matching trajectory vector, and constructing an action space based on the candidate logical field, wherein the historical matching trajectory vector is generated when the preliminary matching of the physical measuring point of the power equipment and the preset standard logical field is performed; calculating the semantic similarity of the physical measuring point of the power equipment and the preset standard logical field; evaluating the unit consistency, the equipment type consistency and the historical matching consistency of the physical measuring point of the power equipment and the preset standard logical field to construct a reward function; constructing the initial decision model based on the state space, the action space and the reward function; performing simulation mapping by using the initial decision model based on the preset strategy optimization algorithm to generate a reward signal, and iteratively optimizing the initial decision model based on the reward signal to obtain the reinforcement learning strategy model, including: performing simulation mapping by using the initial decision model based on the preset strategy optimization algorithm: obtaining the probability corresponding to the selection of the action in the action space from the state in the state space in each simulation mapping execution, generating the reward signal, and ending the simulation mapping; and iteratively optimizing the initial decision model based on the reward signal until the reward signal meets the preset decision requirement to obtain the reinforcement learning strategy model.

[0033] In an embodiment, the initial decision model is constructed based on the semantic vector set and the candidate logical field by using a preset decision algorithm, and the initial decision model is iteratively optimized based on a preset strategy optimization algorithm to obtain a reinforcement learning strategy model, including: constructing a state space based on the semantic vector set, the candidate logical field and a historical matching trajectory vector, and constructing an action space based on the candidate logical field, wherein the historical matching trajectory vector is generated when the preliminary matching of the physical measuring point of the power equipment and the preset standard logical field is performed; calculating the semantic similarity of the physical measuring point of the power equipment and the preset standard logical field; evaluating the unit consistency, the equipment type consistency and the historical matching consistency of the physical measuring point of the power equipment and the preset standard logical field to construct a reward function; constructing the initial decision model based on the state space, the action space and the reward function; performing simulation mapping by using the initial decision model based on the preset strategy optimization algorithm to generate a reward signal, and iteratively optimizing the initial decision model based on the reward signal to obtain the reinforcement learning strategy model, including: performing simulation mapping by using the initial decision model based on the preset strategy optimization algorithm: obtaining the probability corresponding to the selection of the action in the action space from the state in the state space in each simulation mapping execution, generating the reward signal, and ending the simulation mapping; and iteratively optimizing the initial decision model based on the reward signal until the reward signal meets the preset decision requirement to obtain the reinforcement learning strategy model. The semantic vector set of the current to-be-mapped measuring point The semantic vector set of the candidate logical field And the historical matching trajectory vector Together, the semantic vector set, the candidate logical field and the historical matching trajectory vector can comprehensively reflect the current environment perception information; then define the action space A, the action space is a certain item in the current candidate logical field set, that is , which represents mapping the semantic vector of the physical measuring point to a specific logical field Then, the reward function is constructed by considering the semantic similarity score, the consistency of device type and unit, and the consistency of historical matching to evaluate the rationality of the current action, and the expression of the reward function is as follows: ; In the formula, represents the semantic similarity between the semantic vector of the physical measurement point and the semantic vector of the logical field; is a unit consistency index, which is 1 when the units match, and 0 otherwise; is a device type consistency index; represents whether the current action conflicts with the historical mapping strategy; is a hyperparameter used to balance the weights of the rewards.

[0034] Further, an initial decision model is constructed according to the defined state space, action space, and reward function, which is represented as a policy function , which represents the probability of selecting action in state , which is realized by a neural network represented by parameter Finally, the initial decision model is optimized in combination with a preset policy optimization algorithm. In this embodiment, the proximal policy optimization (PPO) algorithm is used as the core reinforcement learning method to learn and optimize the optimal policy. Specifically, in each round of training process, the policy function is repeatedly iterated and optimized by the PPO algorithm. In each iteration and optimization process, the initial decision model is used to select the action corresponding to the probability of the action in the action space from the state in the state space, and the optimal logical data mapping action is selected and filtered out. The logical data mapping action is issued to the physical measurement point for execution, and the generated logical data mapping result is evaluated to calculate the matching accuracy of the logical data mapping result in the semantic space, the consistency index of the unit and the device type , and the comprehensive evaluation value, i.e., the reward signal, is generated in combination with the delay performance and the conflict detection result, and the expression is as follows: ; In the formula, is a weight parameter; is a conflict detection result; then the immediate reward signal is fed back to the policy optimizer to optimize the initial decision model. This mechanism ensures that the parameter update can improve the policy effect and avoid instability caused by large gradients, so that the optimal policy can be learned in continuous interaction with the environment , realize accurate, efficient and adaptive matching of power equipment measuring point and logic field, it is worth mentioning that, in the embodiment, the training stability and strategy generalization ability are improved by using the strategy clipping mechanism, that is, the clipping objective function in the proximal policy optimization (PPO) is adopted: ; In the formula, is the advantage function estimate value, is the clipping threshold, and the mechanism can inhibit the policy update amplitude from being too large, thereby guaranteeing the stability and convergence of the training process.

[0035] In the above scheme, a state space fused with a semantic vector set, candidate logic fields and historical matching trajectories, and an action space based on the candidate logic fields are constructed, a reward function for constraining the semantic similarity between the power equipment physical measuring point and the preset standard logic field, the unit consistency between the power equipment physical measuring point and the preset standard logic field, the device type consistency and the historical matching consistency is designed, an initial decision model is constructed in this way, and then the initial decision model is simulated and mapped in combination with a preset policy optimization algorithm, and iterative optimization is performed according to the generated reward signal, so that a reinforcement learning strategy model is finally obtained. In this way, the static classification problem is converted into a dynamic decision problem considering multiple factors, so as to realize automatic mapping and optimization of the logic field of the physical measuring point, and the adaptability of the power grid system in a complex scene is improved.

[0036] In a preferred scheme, based on the reinforcement learning strategy model, a logic data mapping action meeting the preset requirements is extracted from the candidate logic field, and the logic data mapping action is performed on the to-be-processed physical measuring point of each power equipment to obtain the logic data corresponding to the to-be-processed physical measuring point. The method comprises the following steps: based on the reinforcement learning strategy model, generating logic data mapping action probability distribution data set; based on the logic data mapping action probability distribution data set, extracting the logic data mapping action meeting the preset requirements from the candidate logic field; and based on the logic data mapping action, mapping the to-be-processed physical measuring point of each power equipment to the corresponding logic field to complete the logic data mapping of the power equipment.

[0037] In a preferred scheme, based on the reinforcement learning strategy model, a logic data mapping action meeting the preset requirements is extracted from the candidate logic field, and the logic data mapping action is performed on the to-be-processed physical measuring point of each power equipment to obtain the logic data corresponding to the to-be-processed physical measuring point. The method comprises the following steps: based on the reinforcement learning strategy model, generating logic data mapping action probability distribution data set; based on the logic data mapping action probability distribution data set, extracting the logic data mapping action meeting the preset requirements from the candidate logic field; and based on the logic data mapping action, mapping the to-be-processed physical measuring point of each power equipment to the corresponding logic field to complete the logic data mapping of the power equipment.

[0038] In an implementation of the preferred solution, the trained reinforcement learning model is used to determine the optimal action (equivalent to the preset requirement) from the candidate set of logical fields according to the current policy function The probability distribution of the output is used to extract the optimal (equivalent to the preset requirement) action from the candidate set of logical fields The physical measurement points to be processed are mapped to the corresponding logical fields. Then, for each execution of the power equipment logical data mapping, all action-reward pairs are recorded to an experience buffer for subsequent batch training. To cope with changes in data distribution and fine-tuning requirements of the logical structure during operation, an online small-step incremental learning mechanism is introduced to realize continuous adaptive updating of the policy model, thereby continuously optimizing the logical mapping quality while ensuring system stability. This supports the actual needs of dynamic scalability and accurate semantic modeling in power business scenarios. Specifically, the mapping feedback results are dynamically injected into the new state vector, enabling the next decision to perceive the impact of historical actions on the overall convergence trend. Combining online fine-tuning and incremental learning, the system collects real-time mapping results of newly connected measurement points and injects these data into the experience buffer in small batches. Combined with small-step parameter updating, online incremental learning is realized to ensure continuous adaptive optimization of the model as the operating environment changes, without disrupting the stable strategies learned. A value function baseline correction is used to reduce reward estimation variance, a priority experience replay mechanism is used to improve the utilization of rare samples, and entropy regularization is used to maintain exploration ability and avoid local optima. After the policy model training is complete, the system deploys the learned optimal mapping strategy as the core component of logical data generation and service publishing. The optimal policy model obtained through reinforcement learning training is compiled into a deployable service mapper or lightweight policy engine and integrated into the actual power system environment to realize online execution and intelligent response of logical mapping. The deployed system supports real-time identification and fast semantic analysis of newly connected devices and measurement points, automatically completing the mapping and registration of their logical fields, greatly improving the configuration efficiency and consistency assurance capability of online devices. The deployment stage includes processes such as serialization and loading of the policy model, integration and deployment of the inference engine, encapsulation and development of interfaces and middleware, ensuring efficient operation and fast response of the mapping process in the actual business system. At the same time, the power system supports online inference mechanism, enabling fast semantic analysis and logical field matching of newly connected measurement points without the need for retraining. It is worth mentioning that at the data service level, the system further registers the mapped logical fields to the service publishing module, supporting the provision of structured data access interfaces, realizing unified and standardized logical data access services, and providing reliable data support for upper-layer businesses such as dispatching, operation and maintenance, and monitoring.

[0039] In the above scheme, the probability distribution of logical data mapping actions is obtained by using the iteratively optimized reinforcement learning strategy model. Based on this probability distribution, logical data mapping actions that meet the preset requirements are selected and then sent to the physical measurement points of the power equipment. The physical measurement points of the power equipment execute the corresponding logical data mapping actions, mapping the physical measurement points of each power equipment to the corresponding logical fields to complete the logical data mapping of the power equipment. An incremental learning mechanism is also introduced to feed back the mapping results generated after each execution of the logical data mapping action to a preset experience buffer pool with a preset step size. This allows for the adjustment of the logical data mapping action probability distribution dataset or the fine-tuning of the logical structure of the reinforcement learning strategy model. This enables continuous optimization of the reinforcement learning strategy model, which helps to achieve automatic mapping and optimization of the logical fields of physical measurement points and improves the adaptability of the power grid system in complex scenarios.

[0040] Example 2 See Figure 2 , Figure 2 This is a schematic diagram of the module structure of a power equipment logic data mapping system according to a certain embodiment of the present invention. Figure 2 As shown in the figure, this embodiment of the invention also provides a power equipment logical data mapping system, including: a semantic attribute construction module 201, a semantic analysis module 202, a reinforcement learning strategy model construction module 203, and a logical data mapping module 204; the semantic attribute construction module 201 is used to construct a semantic attribute matrix based on pre-collected power equipment physical measurement point data and their label information; the semantic analysis module 202 is used to construct a semantic vector set and candidate logical fields based on a preset semantic analysis algorithm and the semantic attribute matrix; the reinforcement learning strategy model construction module 203 is used to construct an initial decision model based on the semantic vector set and candidate logical fields through a preset decision algorithm, and iteratively optimize the initial decision model based on a strategy optimization algorithm to obtain a reinforcement learning strategy model; the logical data mapping module 204 is used to extract logical data mapping actions that meet preset requirements from the candidate logical fields based on the reinforcement learning strategy model, and execute logical data mapping actions on the physical measurement points to be processed of each power equipment to obtain the logical data corresponding to the physical measurement points to be processed.

[0041] The embodiment of the present application provides a power equipment logical data mapping system, physical measuring point data and label information of power equipment are collected to perform structured modeling, semantic attribute matrix, semantic vector set and candidate logical field are obtained, a preset decision algorithm and a strategy optimization algorithm are introduced, a reinforcement learning strategy model is constructed, and it is guaranteed that the reinforcement learning strategy model can continuously optimize and update mapping strategies, so that the reinforcement learning strategy model can automatically extract logical data mapping actions meeting preset requirements from the candidate logical field, and the logical data mapping actions are performed on the physical measuring points to be processed of each power equipment, and finally, the logical data mapping of the power equipment is completed. Therefore, the deep meaning of the measuring point label is understood through semantic analysis, and the reinforcement learning model is constructed by using structured information, so that the logical field of the physical measuring point is automatically mapped and optimized, and the adaptability of the power grid system in a complex scene is improved.

[0042] A power equipment logical data mapping terminal, the terminal comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, the processor implementing a power equipment logical data mapping method when executing the computer program.

[0043] A power equipment logical data mapping storage medium, the storage medium comprising a stored computer program, wherein the device where the storage medium is located performs a power equipment logical data mapping method when the computer program runs.

[0044] The above only describes the preferred embodiments of the present application, and it should be pointed out that, for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and modifications can be made, and these improvements and modifications should also be considered as the protection scope of the present application.

[0045] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in combination with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any suitable way in any one or more embodiments or examples. In addition, the skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0046] In addition, the terms "first", "second" are only for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one feature. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

Claims

1. A method for mapping logical data of power equipment, characterized in that, include: Based on the pre-collected physical measurement point data of power equipment and its tag information, a semantic attribute matrix is ​​constructed; Based on the preset semantic analysis algorithm and the semantic attribute matrix, a semantic vector set and candidate logical fields are constructed; Based on the semantic vector set and the candidate logical fields, an initial decision model is constructed using a preset decision algorithm, and the initial decision model is iteratively optimized using a policy optimization algorithm to obtain a reinforcement learning policy model. Based on the reinforcement learning strategy model, logical data mapping actions that meet preset requirements are extracted from the candidate logical fields, and the logical data mapping actions are executed on the physical measurement points to be processed of each power equipment to obtain the logical data corresponding to the physical measurement points to be processed.

2. The power equipment logic data mapping method as described in claim 1, characterized in that, The semantic attribute matrix is ​​constructed based on the pre-collected physical measurement point data of power equipment and its tag information, including: Collect physical measurement point data and their label information of power equipment to construct a set of observation variables; Based on the physical affiliation and logical function positioning of each power equipment physical measurement point, a mapping set is constructed, and the mapping relationship corresponding to each power equipment physical measurement point is constructed. Based on the mapping relationship between the physical measurement points of each power equipment, a numerical representation of each physical measurement point of the power equipment in several dimensions is constructed to obtain a semantic attribute matrix.

3. The power equipment logic data mapping method as described in claim 1, characterized in that, The construction of a semantic vector set and candidate logical fields based on a preset semantic analysis algorithm and the semantic attribute matrix includes: The semantic attribute matrix is ​​converted into low-dimensional continuous semantic vectors by a preset semantic analysis algorithm to obtain a semantic vector set; Based on a preset similarity measurement algorithm, the semantic distance between different physical measurement points of power equipment in the semantic vector set or the semantic distance between different physical measurement points of power equipment and preset standard logical fields is calculated to construct a semantic space; Based on a preset search algorithm, physical measurement points of power equipment are initially matched with preset standard logical fields in the semantic space to obtain candidate logical fields.

4. The power equipment logic data mapping method as described in claim 3, characterized in that, Based on the semantic vector set and the candidate logical fields, an initial decision model is constructed using a preset decision algorithm, and the initial decision model is iteratively optimized using a policy optimization algorithm to obtain a reinforcement learning policy model, including: Based on the semantic vector set, the candidate logical fields, and the historical matching trajectory vectors, a state space is constructed, and based on the candidate logical fields, an action space is constructed. The historical matching trajectory vectors are generated when the physical measurement points of power equipment are initially matched with the preset standard logical fields. Calculate the semantic similarity between the physical measurement points of the power equipment and the preset standard logical fields; Evaluate the consistency of the physical measurement points of the power equipment with the preset standard logical fields in terms of unit, equipment type, and historical matching, and construct a reward function; Based on the state space, the action space, and the reward function, an initial decision model is constructed. Based on a preset strategy optimization algorithm, a simulation mapping is performed through the initial decision model to generate a reward signal, and the initial decision model is iteratively optimized according to the reward signal to obtain a reinforcement learning strategy model.

5. The power equipment logic data mapping method as described in claim 4, characterized in that, The optimization algorithm based on a preset strategy involves performing simulated mapping through the initial decision model to generate a reward signal, and iteratively optimizing the initial decision model based on the reward signal to obtain a reinforcement learning policy model, including: The simulation mapping is performed by optimizing the algorithm with a preset strategy and using the initial decision model: obtaining the probability of selecting an action in the action space from the state in the state space for each simulation mapping execution, generating a reward signal, and ending the simulation mapping. Based on the reward signal, the initial decision model is iteratively optimized until the reward signal meets the preset decision requirements, thereby obtaining a reinforcement learning policy model.

6. The power equipment logic data mapping method as described in claim 1, characterized in that, Based on the reinforcement learning policy model, logical data mapping actions that meet preset requirements are extracted from the candidate logical fields. These actions are then executed on the physical measurement points of each power device to obtain the logical data corresponding to each physical measurement point, including: Based on the reinforcement learning policy model, a logical data mapping action probability distribution dataset is generated. Based on the logical data mapping action probability distribution dataset, logical data mapping actions that meet preset requirements are extracted from candidate logical fields; Based on the aforementioned logical data mapping action, the physical measurement points to be processed for each power device are mapped to the corresponding logical fields to complete the logical data mapping of the power devices.

7. The power equipment logic data mapping method as described in claim 6, characterized in that, Each time the reinforcement learning policy model is executed, a logical data mapping action that meets preset requirements is extracted from the candidate logical fields. After the logical data mapping action is executed on the physical measurement points to be processed for each power device to obtain the logical data corresponding to the physical measurement points to be processed, the process further includes: Receive several mapping results obtained by each physical measurement point of each power device performing the logical data mapping action, and inject several mapping results into a preset experience buffer pool with a preset step size. Based on the preset experience buffer pool, the reinforcement learning strategy model is incrementally learned, and the model is adaptively fine-tuned according to the logical data mapping action probability distribution dataset or logical structure to complete the model optimization.

8. A logic data mapping system for power equipment, characterized in that, Performing a power equipment logic data mapping method as described in any one of claims 1 to 7, comprising: Semantic attribute construction module, semantic analysis module, reinforcement learning policy model construction module, and logical data mapping module; The semantic attribute construction module is used to construct a semantic attribute matrix based on the pre-collected physical measurement point data of power equipment and its tag information; The semantic analysis module is used to construct a semantic vector set and candidate logical fields based on a preset semantic analysis algorithm and the semantic attribute matrix; The reinforcement learning policy model construction module is used to construct an initial decision model based on the semantic vector set and the candidate logical fields using a preset decision algorithm, and to iteratively optimize the initial decision model based on a policy optimization algorithm to obtain a reinforcement learning policy model. The logical data mapping module is used to extract logical data mapping actions that meet preset requirements from the candidate logical fields based on the reinforcement learning strategy model, and to execute the logical data mapping actions on the physical measurement points to be processed of each power equipment to obtain the logical data corresponding to the physical measurement points to be processed.

9. A logic data mapping terminal for power equipment, characterized in that, The terminal includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a power equipment logic data mapping method as described in any one of claims 1 to 7.

10. A logic data mapping storage medium for power equipment, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform a power equipment logic data mapping method as described in any one of claims 1 to 7.

Images